【摘要】 针对粗粒度的商品评论情感分析不能详尽地提供用户喜好问题,提出一种基于支持向量机(SVM)结合点互信息(PMI)的细粒度商品评论情感分析方法。首先,使用卡方检验方法进行文本特征选择和降维;接着,对朴素贝叶斯、决策树、支持向量机(SVM)、K最邻近算法(KNN)四种常用情感分类方法进行比较,支持向量机(SVM)的召回率和精确率最高,均达到94.5%,所以使用支持向量机(SVM)对商品评论进行粗粒度的情感分析;然后,根据人工经验总结典型的商品属性,使用点互信息(PMI)方法对商品属性扩充;最后针,对扩充后的商品属性,在以上粗粒度的商品评论情感分析基础上,进行细粒度的情感分析及统计。细粒度的商品评论情感分析,可使厂家看到用户对产品属性的喜好,以及在产品设计、销售及服务中需要改进的方面。更多还原

【Abstract】 Aiming at the problem that coarse-grained commodity review sentiment analysis cannot provide the user preference in detail, a method of fine-grained commodity comment sentiment analysis based on Support Vector Machine(SVM) combined with Point Mutual Information(PMI) was proposed. Firstly, chi-square test method was used for text feature selection and dimensionality reduction. Then, four common sentiment classification methods, such as Naive Bayes, decision tree, Support Vector Machine(SVM) and K Nearest Neighbors(KNN) algorithm, were compared. SVM had the highest recall rate and accuracy rate of 94.5%, therefore, it was used to conduct coarse-grained sentiment analysis of commodity reviews. Then, based on manual experience to summarize typical commodity attributes, PMI was used to expand the product attributes. Finally, based on the expanded product attributes and the above-mentioned coarse-grained product review sentiment analysis, fine-grained sentiment analysis and statistics was performed. Fine-grained product reviews sentiment analysis allows manufacturers to observe user preferences for product attributes and areas for improvement in product design, sales, and service.更多还原